Sequential and Parallel Hybrid Approaches of Augmented Neural Networks and GRASP for the 0-1 Multidimensional Knapsack Problem

Abstract

There are a lot of problems, whose solutions are based on 0-1 multidimensional problem. Since this combinatorial optimization problem is \(\mathcal {NP}\)-hard, one of the approaches to solve it is the use of metaheuristics. For this problem, even heuristics consume a lot of time to find a solution, which motivates the search for alternatives capable of making the use of such techniques less time-consuming. Among these alternatives, the use of parallelization strategies deserves to be highlighted, once it may lead to reduced execution times and/or better quality results. In this work we propose a hybrid approach of augmented neural networks and GRASP for the 0-1 multidimensional knapsack problem, we describe a sequential and a GPGPU implementation and measure the achieved speedups. We also compare our results with the ones obtained by other metaheuristics. The obtained results show that the proposed approach can achieve better quality solutions than some of the other algorithms found in the literature. These solutions can lead to better solutions to real problems that can be modeled with the 0-1 MKP.